It’s 4:45 p.m. on Friday. A client wants a lease abstract and a five-year pro forma by Monday — and the tenant rep just emailed a 90-page lease with four amendments. A few years ago, you’d have canceled your weekend plans to spend two days highlighting rent escalations and typing dates into a spreadsheet.
Today, an analyst can drop the whole document into an AI chatbot, which generates a draft abstract in minutes.
Sounds great, right? As AI use has increased in property management, it also raises privacy and accuracy risks. In IREM and AppFolio’s survey of 3,662 U.S. property management and real estate professionals, 58% reported using AI in their role, and among those users, 79% used general-purpose assistants.
Uploaded data matters more than speed because a lease is one of the most information-dense, confidential documents in commercial real estate.
What info does a lease contain?
One lease contains dozens of data points and sensitive information: base rent, escalations, percentage rent, sales breakpoints, TI allowances, guarantees, exclusivity clauses, sublease and assignment rights, renewal options with notice windows, expense recovery caps, and termination rights.
Add amendments, side letters, and estoppels, and the document becomes a roadmap showing how the parties negotiated the deal and what the tenant or landlord can do.
What happens when that document leaves your company’s control? Many leases include confidentiality provisions; client engagement letters and NDAs also may restrict who can process documents and how. Uploading a lease to an unapproved tool can quickly become a contractual concern.
What AI tools do with your data
Many people misunderstand the difference between free and paid AI tools.
Data handling depends on the product, account type, and settings. Some personal AI services will use your conversations and files to improve their models unless you specifically opt out. Consumer ChatGPT, for example, enables data sharing by default in personal workspaces. Other consumer products offer a model-improvement setting that, when toggled on, can retain de-identified data for years. (Anthropic’s privacy center says it can keep training-eligible chats for up to five years.)
Meanwhile, many business and enterprise offerings exclude organizational data from model training by default and add administrator controls. They currently include:
- ChatGPT’s business plans
- Anthropic’s commercial products (Claude for Work)
- Microsoft’s organizational Copilot
- Gemini for Google Workspace
A vocabulary note: An enterprise subscription describes contractual and administrative protections, whereas a private or self-hosted deployment describes how a system operates. The two are not interchangeable.
Two caveats apply for CRE professionals.
First, “not used for training” isn’t the same as “not stored,” “never reviewed,” or “safe to upload.” Consumer plans may retain conversations on their servers, providers may review flagged content for safety reasons, and retention windows and deletion rules vary by product and plan. The specifics differ even among the enterprise products themselves:
- Anthropic’s API deletes inputs and outputs within 30 days.
- Microsoft stores organizational Copilot interactions in each user’s mailbox under admin-configured retention rules.
- Google keeps Gemini in Workspace prompts and responses for 90 days to indefinitely, depending on the admin settings.
If you plan to use AI to help streamline your lease administration and workflow, find out how long the system keeps the uploaded lease and who can access it. Don’t assume it disappears automatically after a set time. Read the fine print.
Second, these policies change frequently. What was true about an AI tool’s data handling this year (or even this quarter) may change. Firms planning to adopt any AI tool should read the vendor’s current data-use policy, date their review, and set a reminder to review it again every 3-6 months.
Banning AI isn’t a practical policy
Some CRE firms prohibit AI outright. A written prohibition alone doesn’t guarantee people will stop using it without authorization.
A 2026 survey of 1,250 office professionals in non-IT roles at companies with $500 million or more in revenue found that 66% had used AI tools at work even while believing their company’s policies didn’t allow it. Nearly 90% (88%) said they’d shared work-related information with public AI tools, and about 33% admitted sharing financial information or confidential company documents and strategies. While the survey wasn’t CRE-specific, it shows employees use AI even though they believe company policy prohibits it.
The takeaway isn’t that employees are reckless. It’s the policy that’s hard to enforce — but there is a different answer than a total ban. Enforcing a total ban is virtually impossible (think: closing the barn door after the horse has left), but firms have another option.
An AI governance policy appropriate for a CRE firm
The alternative to banning AI? Governing it. You don’t have to start from scratch, because standards bodies, including those within the CRE sector, have already drafted recommendations.
The Royal Institution of Chartered Surveyors (RICS), whose members and regulated firms work across CRE worldwide, published the first global professional standard for the responsible use of AI in surveying practice, mandatory for RICS members and regulated firms since March 2026.
The standard applies where AI use materially affects the delivery of surveying services, and it explicitly addresses confidentiality. Firms must safeguard private and confidential data and refrain from uploading it to AI systems unless they have express written consent from affected stakeholders in advance and have taken reasonable steps to ensure the upload poses no unacceptable risk.
The standard also requires at least annual staff training on the privacy and confidentiality risks AI poses. For firms outside RICS regulation, these requirements offer a useful benchmark.
Legal sector guidance offers a parallel. The American Bar Association’s Formal Opinion 512 interprets lawyers’ confidentiality duties and calls for informed client consent before entering representation-related information into self-learning AI tools that pose the disclosure risks discussed in the opinion.
Most firms can start by incorporating these items into their AI policy:
- Approved tools: Name the specific AI platforms and account types staff may use (ideally, firm-managed business or enterprise accounts).
- Permitted data: Define what staff can and cannot upload. Before client documents go anywhere, check client instructions, confidentiality commitments, and organizational approval; obtain client consent where required.
- An owner: Designate one person or committee responsible for vetting tools, updating policy, and answering questions.
- Human review: Require that any AI-generated abstract, rent schedule, or pro forma be verified against the source documents by a qualified professional before sharing with clients.
- A risk management protocol: Provide detailed guidelines about what staff should do if unauthorized data is uploaded (report it immediately, no blame games). A staff member who hides a mistake is a bigger risk than one who made a mistake and owned it.
- A review schedule: Revisit the policy every six months, as the tools and their policies change regularly.
For further reference points, the U.S. National Institute of Standards and Technology (NIST) publishes a voluntary AI risk management framework. An international standard (ISO/IEC 42001) covers AI management systems. The accounting profession has also published practical guidance on drafting AI policies. Review these frameworks to create one that works for your purposes.
Keep a human in the loop
Another — and arguably the most important — caveat: even the best AI tool that’s fed the most detailed prompt can get a lease wrong.
The draft your tool generates in minutes is a starting point. Before you put anything into a client deliverable, give it a thorough review of all the information: critical dates, option windows, escalation math, expense recoveries, amendment dates, expense caps, etc. Read carefully for content that sounds plausible — like a renewal option — but isn’t in the lease. AI can shorten drafting, but you cannot skip the professional review.
The bigger picture
Leases aren’t the only documents at risk. Loan files, appraisals, and purchase agreements raise the same confidentiality questions, and the same answer applies: use approved tools, know what you’re uploading, and review the output before sharing with a client.
Are you a commercial real estate investor or seeking a specific property to meet your company’s needs? We invite you to talk to the professionals at CREA United, an organization of CRE professionals from over 65 firms representing all disciplines within the CRE industry, from brokers to subcontractors, financial services to security systems, interior designers to architects, movers to IT, and more.
